Tacta wires up factory workers to teach robot hands how to touch

Tacta Systems has come out of stealth with TactaBot, an integrated platform built around three pieces: the Tacta Hand, a Skill Capture data-collection system, and a Dexterous Intelligence model that connects them. The company raised USD 75 million from investors including SoftBank, Woven Capital, B Capital, Matter Venture Partners and America's Frontier Fund. It was founded by Andreas Bibl, now executive chairman, and chief executive Vikram Pavate. The hand itself is human-scale, with five fingers (a three-finger variant is also offered) and 15 independently actuated joints, three per finger. Actuation uses a proprietary approach the company calls Fluidic Tendon, pushing pressurised fluid through cable runs rather than packing motors into the finger. Peak fingertip force is 25 newtons, roughly 2.5 kilograms, and Tacta quotes durability of 30 million operating cycles. What sets the announcement apart is where the training data comes from. Most attempts to give robots fine manipulation skills rely on teleoperation: an engineer drives the robot through a task, and the robot learns from its own recorded motion. Tacta instead instruments the human. The Tacta Glove carries 256 tactile sensors per fingertip and records force, motion, temperature and video simultaneously while a skilled worker does the job at their normal pace, streaming the data wirelessly. The resulting dataset trains a model pretrained on general tactile skills and then fine-tuned on a specific customer's task. That approach depends on sensors small enough to sit on a fingertip without changing how the hand works. Tacta built its own: a non-MEMS design smaller than a grain of sand, spaced at a one-millimetre pitch, with analogue-to-digital conversion integrated into the sensor. The stated pressure range runs from 250 pascals to 700,000 pascals, sampling 400 times a second, with temperature resolution of a tenth of a degree. One detail in the specification is unusually candid. Tacta classifies the fingertip sensors as consumables and designs them to be swapped out, with servicing measured in minutes. Touch sensors wear down under the abrasion of real assembly work, and treating that as a scheduled replacement rather than a failure is a concession most dexterity demonstrations avoid making in public. The target work is the kind that has resisted automation for decades: seating connectors, assembling wire harnesses, precision electronics assembly. Tacta says first deployments with manufacturers in electronics, AI infrastructure and automotive are planned for early 2027, which means the claims remain unverified outside the company's own demonstrations for now.